Support Polars DataFrame export from QueryResult

[alexander-beedie]

Closes #1662 (allow export to Polars as well as Pandas).

Hi, thought this might be helpful; I'm one of the Polars devs, so took a look at it. As @pankajp mentioned, Polars can zero-copy init from Arrow, so very little new code was required to enable this. Instead I put most of the focus into generalising the existing Arrow unit tests so that they can also validate the Polars results, ensuring good code coverage. Hope this is welcome!

FYI: I'm also enabling kuzu support on the Polars side (via our read_database init method); there's a nice symmetry (and broader discoverability) in having the functionality available in both APIs 👌 (See: pola-rs/polars#14822, which was just merged).

(Looking at kuzu I think there may be some good use-cases for it at work too, so I'm going to turn a few people on to this; thanks for the great work :)

[mewim]

LGTM. The CI will not pass because we used a few internal secrets in our tests which external contributors do not have access to. We might have to reorder the CI pipeline to make sure that at least the Python tests pass on the platforms we support to merge this.

[semihsalihoglu-uw]

Thanks a lot for the contribution @alexander-beedie. Happy to chat with you and your colleagues if you want to use Kuzu at your work. Please reach out to us over Discord.

[prrao87]

These changes look great! Thanks a lot for the PR!

[prrao87]

@alexander-beedie I took a look at the ConnectionExecutor and it looks great! Once it's in the next release I'll do a thread describing its use and here's looking forward to bringing more Polars users to Kùzu and vice versa :)

[alexander-beedie]

@alexander-beedie I took a look at the ConnectionExecutor and it looks great! Once it's in the next release I'll do a thread describing its use and here's looking forward to bringing more Polars users to Kùzu and vice versa :)

Sounds good to me; was one of our users that pointed me in this direction, and it looks really promising. I'm sure there will be some use-cases where clean interop between the two will lead to some powerful pipelines/capabilities 😎